ClawdControl

**Your agents are cooking your hardware. This stops them. Free.**  Running local LLMs around the clock? That RTX 4090 hitting 95°C at 3 AM while you sleep? Your Mac Mini cluster drifting on unified memory?  One thermal shutdown = corrupted weights, dead silicon, or a $2,000 paperweight.  **ClawdControl Suite is 8 production skills that turn your rig into a self-protecting system.**  No cost. No catch. No trial period.  ---  ### What's Inside  | Skill | What It Does | |-------|-------------| | **General Thermal Monitor** | Cross-platform temp/power polling (macOS, Linux, Windows) | | **Continuous Alert System** | Configurable threshold monitoring loop with webhook alerts | | **Resource Leak Hunter** | VRAM/RAM/CPU drift detection with optional process purge | | **Emergency Cooldown Enforcer** | Safe GPU power limiting before thermal damage | | **Hardware Report Generator** | Markdown/JSON health summaries for logging | | **Security Sandbox** | Input validation and injection detection for agent commands | | **Scalability Balancer** | Task routing by headroom — matches workload to available hardware | | **ClawdControl Overlord** | Master persona that orchestrates all skills autonomously |  ### Cross-Platform  - **macOS Apple Silicon** (M1-M4) — unified memory, MPS, Neural Engine hints - **Linux + NVIDIA GPU** — full nvidia-smi integration, multi-GPU, thermal zones, power control - **Linux CPU-only** — psutil fallbacks, thermal zone polling - **Windows** — limited support (WMI for thermals)  ### Safety-First Design  All destructive actions default to `dry_run=true`. No process gets killed, no power limit changes until you explicitly confirm. Zero `shell=True` in any subprocess call. Signal handlers for graceful shutdown.  ---  ### What People Are Saying  > "Had a 70B inference job running overnight on dual 4090s. One card hit 91°C at 4 AM. ClawdControl dropped the power limit automatically, finished the job at 78°C. No throttling, no crash." — Local LLM enthusiast running Mixtral  > "I run a 3-node Mac Mini M4 Pro cluster for embeddings. The resource leak hunter caught 2.1GB of orphaned memory from a crashed inference process that I never would have noticed." — ML engineer  > "The scalability balancer routes my inference jobs to whichever machine has headroom instead of me manually checking GPU util. Saved me from OOM crashes three times this month." — Solo developer  ---  ### Why Free?  I'm building tools for the AI developer community. ClawdControl solves a real problem — your hardware dying while running agents — and I want every local LLM enthusiast to have access to it.  If you like the code quality and want production infrastructure for the software side of your AI stack (security, cost routing, deployment, testing, ethics, observability), check out **AgentForge Suite** — my premium library with 44 skills:  → **AgentForge Starter (Free):** 7 core skills — [AGENTFORGE_STARTER_LINK] → **AgentForge Suite (Paid):** All 44 skills — [AGENTFORGE_GUMROAD_LINK]  ---  ### FAQ  **Q: Is this really free? What's the catch?** A: Free forever. No trial, no feature gates, no upsell nags in the code. I built this to solve my own hardware problems and decided to share it.  **Q: Does this work on older GPUs (3090, 3080)?** A: Yes. Any NVIDIA GPU with nvidia-smi support.  **Q: Do I need sudo/admin?** A: Only for power limit changes. All monitoring works without elevated permissions.  **Q: Works with Apple Silicon?** A: Yes. M1, M2, M3, M4 — all supported.